In this paper a modern statistical technique of multivariate analysis is applied to an indoor radon concentration data base. Several parameters are more or less significant in determining the radon concentration inside a building. The elaboration of the information available on South Tyrol makes it possible both to identify the statistically significant variables and to build up a statistical model that allows us to forecast the radon concentration in dwellings, when the values of the same variables involved are given. The results confirm the complexity of the phenomenon.